13:00:56 <baoli> #startmeeting PCI passthrough 13:00:57 <openstack> Meeting started Tue Oct 28 13:00:56 2014 UTC and is due to finish in 60 minutes. The chair is baoli. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:00:58 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:01:00 <openstack> The meeting name has been set to 'pci_passthrough' 13:01:04 <baoli> Hi there 13:01:15 <irenab> hi 13:01:59 <baoli> updated the meeting wiki https://wiki.openstack.org/wiki/Meetings/Passthrough#Agenda_on_Oct._28th.2C_2014 13:03:48 <sadasu> Hi 13:04:04 <beagles> hi 13:04:09 <irenab> baoli: Thanks! 13:04:35 <baoli> reposting https://wiki.openstack.org/wiki/Meetings/Passthrough#Agenda_on_Oct._28th.2C_2014 13:05:30 <baoli> I started tagging sr-iov/pci-passthrough related bugs with pci-passthrough 13:06:08 <beagles> baoli++ 13:06:29 <sadasu> baoli: started looking at your patches but didn't get through them 13:06:57 <baoli> Please tag bugs with pci-passthrough as you see fit. 13:07:11 <irenab> baoli: I'll use the same on neutron side as well 13:07:31 <sadasu> is there a Nova session on current pci-passthrough changes at the summit? 13:07:49 <baoli> there are two links: one for nova, one for neutron, to find bugs with that tag 13:08:07 <baoli> sadasu, I just checked but didn't see one 13:09:27 <irenab> I do not think there is 13:10:10 <irenab> I think we can meet at pod to discuss 13:11:02 <irenab> summit sessions ae more for talks to reach consensus and not update on progress 13:13:19 <baoli> there are a bunch of Cross-project workshop, but they are not finalized yet 13:13:54 <sadasu> sorry, my connection flapped 13:13:59 <sadasu> did I miss anything? 13:14:53 <sadasu> test 13:14:58 <baoli> sadasu, not really. no talk scheduled for pci-passthrough at the summit. We're saying if we want to meet at pod 13:15:21 <sadasu> ok. +1 for meeting at the pod 13:15:51 <baoli> sometime on Monday? 13:16:11 <irenab> works for me 13:16:55 <irenab> need to see if there is some schedule for pod time sharing 13:18:00 <sadasu> Monday morning or afternoon? 13:18:34 <irenab> Let's send proposal over emails, once finalize the schedule 13:18:54 <baoli> that works. 13:19:04 <sadasu> irenab: ok. when are you getthing there? 13:20:21 <irenab> I am joing the upsteam training, so will be there over weekend 13:21:33 <irenab> sadasu: when do you arrive? 13:22:24 <sadasu> Sunday morning 13:23:17 <irenab> sadasu: cool, we can meet at Sunday evening 13:23:38 <irenab> baoli: there is recent ml thread regarding pci whitelist format, do we want to discuss it here? 13:24:01 <baoli> irenab, sure 13:24:22 <baoli> but I'm not clear what it is breaking now 13:24:51 <beagles> irenab, you are referring to sdague's comments? 13:25:08 <irenab> beagles: yes 13:26:09 <irenab> He suggests quite reasonable solution with additional file, but I remember when we proposed it initially it has some objections 13:26:46 <beagles> yeah... I was always under the impression that the current pci whitelist was a sort of interim solution - with some folks wanting a more centrally manageble configuration. 13:27:05 <irenab> beagles: I agree 13:27:31 <beagles> so maybe we should be looking in that direction instead. As an admin and deployment thing, I would expect the NFV crowd having some thoughts on this. 13:28:56 <irenab> beagles: do you have any idea how they want it to be managed? 13:29:24 <irenab> API, auto-discovery, other.... 13:29:29 <beagles> irenab, no idea. 13:30:08 <beagles> irenab, I'm trying to recall who mentioned it in Atlanta .. Adrian maybe? 13:30:34 <sadasu> yeah, from the emails, i am only able to gather what is not preferred... 13:30:56 <beagles> I'll fire a reply to the ML about that whitelist being an interim solution if that's okay with everyone 13:31:06 <irenab> beagles: +1 13:31:52 <baoli> Initially I had the impression that Sean doesn't like how it's configured through devstack. But it seems he intended to change how it's configured in nova. Right? 13:32:23 <irenab> baoli: I think so 13:32:29 <beagles> seems that way 13:32:49 <sadasu> baoli: from the emails..it does seem that way 13:34:03 <irenab> is there any specific topic we want to discuss at the summit? maybe worth to advertise once we set a time and location for others to join 13:34:06 <sadasu> But, I am not sure if having another file for the pci passthrough config and giving a pointer to that in nova config would be a widely accepted solution either 13:34:27 <baoli> Regarding whether or not it's an interim solution, keep in mind that's how it was from the very beginning of PCI passthrough. 13:34:33 <irenab> I think beagles is right, we need to provide other tool ro set it 13:34:45 <sadasu> baoli: agreed 13:35:11 <baoli> but there might be better way to facilitate the user to configure it. 13:35:30 <sadasu> irenab: this would be aperfect topic to nail down at the summit 13:36:35 <irenab> sadsu7: agree, I think we need to have some proposals to discuss it. Hopefuly beagles' mail will raise ideas 13:37:10 <sadasu> ok 13:37:11 <baoli> look forward to it. 13:38:28 <sadasu> should we involve any oslo folks in the discussion for ideas? 13:38:58 <irenab> baoli: do you want to go over new bugs? 13:39:27 <baoli> irenab, we can look at the first bug on the list 13:41:03 <baoli> any thoughts? It's actually already found long back and there are existing patches. 13:42:11 <baoli> It's kind of critical for restart 13:42:39 <irenab> baoli: seems there is afix lready submitted 13:42:56 <irenab> https://review.openstack.org/#/c/131321/ 13:44:14 <baoli> missed that earlier. 13:45:17 <irenab> I'll try to verify the fix 13:45:31 <baoli> well, that won't fix the fundamental issue. 13:45:48 <irenab> baoli: can you please summarize the initial issue 13:46:37 <sadasu> fix seems simple..but there is no explation of the root cause of the problem 13:46:57 <baoli> irenab, during restart, the compute node needs to reconcile with the DB in terms of the PCI device records. 13:47:23 <baoli> That's not done properly. 13:48:17 <baoli> https://review.openstack.org/#/c/82206/ https://review.openstack.org/#/c/102298/ attempted to fix that. 13:48:55 <irenab> baoli: do you mean DB should be in sync with node current PCI devices? 13:49:44 <baoli> irenab, nova-compute needs to sync up with the DB during its restart 13:51:04 <irenab> baoli: to reflect current compute devices and update device availability state? 13:51:24 <baoli> irenab, that's right. 13:51:44 <baoli> in restart case, all the VMs keep running on the node (unless the node itself gets restarted) 13:52:03 <irenab> baoli: but this is not the case of this bug, right? 13:52:20 <baoli> irenab, it is. 13:52:21 <irenab> if node restarts, devices may be changed 13:53:05 <irenab> by the way, what happens in case the white list is changed and then nova is restarted 13:53:15 <baoli> irenab, I wouldn't think admin would remove, but may add new. Most use case would be to upgrade 13:53:35 <baoli> sorry, wouldn't remove any existing 13:54:41 <irenab> so actually the case to fix, is nova restart with same or extended white list settings while there are running VMs with allocated PCI devices 13:54:49 <sadasu> irenab: the new whitelist should be used for VMs that are freshly brought up after the whitelist change 13:55:01 <sadasu> existing VMs should continue as is 13:55:23 <baoli> irenab, currently even without any change, it's not working as designed, or the code has never worked for the restart case. 13:55:33 <baoli> without any change of WL 13:55:35 <irenab> sadsu: yes, I was afraid of the case, where white list maybe changed in the way that it won't contain allocated devices 13:55:36 <sadasu> baoli: the 2 review links you posted are changing the same files 13:55:48 <sadasu> are they 2 diff fixes that you are proposing? 13:56:36 <irenab> baoli: do you know if Heyongli fixing this issue? 13:56:37 <sadasu> irenab: agreed..I knew that you were referring to that case...but too bad we can't change what is already allocated correct? 13:56:51 <baoli> sadasu, kind of. 13:57:22 <baoli> irenab, I need to take a close look at it. 13:57:58 <baoli> But my first impression is that it doesn't address the issue completely 13:58:09 <irenab> baoli: it is really good idea to track this issues at group wiki page 13:58:45 <baoli> irenab, agreed. 13:59:19 <baoli> it's listed as the top bug. 13:59:31 <irenab> I'll try to take a look at this issue as well 13:59:57 <baoli> irenab, great 14:00:15 <baoli> I think that the time is up, we can continue at the summit 14:00:22 <baoli> See you guys at the summit. 14:00:30 <irenab> guys and girls, see you in Paris 14:00:59 <baoli> #endmeeting